Professional Documents
Culture Documents
Technical Guide 2, Mar 08 PDF
Technical Guide 2, Mar 08 PDF
T e c h n i c a l Gu i d e 2
A Guide on Measurement Uncertainty in
Chemical & Microbiological Analysis
The Singapore Accreditation Council (SAC) would like to thank members of the Task Force
on Measurement Uncertainty for Chemical & Microbiological Analysis and their respective
organization for their effort & contributions in establishing this technical guide.
Name Organisation
SAC would also like to thank the following laboratories for contributing additional worked
examples to the second edition of the Technical Guide 2:
Note: Use of the material does not imply equivalence with the EURACHEM/CITAC guide.
5. Evaluation Methods 4
8. Reporting Uncertainty 17
9. General Remarks 18
1.1 The International Standard ISO/IEC 17025:2005 on "General Requirements for the
Competence of Testing and Calibration Laboratories" [1] has included a series of
clauses on the estimation of measurement uncertainty for calibration and testing
laboratories. It requests the assessment of uncertainty of test results during method
validation and requires testing laboratories to have and apply procedures for
estimating uncertainties of measurement in all test methods except when the test
methods preclude such rigorous estimations.
1.2 The SAC-SINGLAS 002 document on "General Requirements for the Competence
of Calibration and Testing Laboratories" [2] also states that a laboratory shall use
appropriate methods and procedures, including an estimation of uncertainty in all
measurements, and indicate the quantitative results accompanying with a statement
of the estimated uncertainty.
1.4 It is the aim of this Guide to give general information of the application of uncertainty
to chemical analysis and microbiological analysis and its effects on compliance. This
Guide outlines the current thinking of methodology, based on the methods
prescribed in the ISO Technical Advisory Group on Metrology (TAG4’s) lengthy
document ISO guide 98 on "Guide to the Expression of Uncertainty in
Measurement", commonly known as GUM [4] in 1995, the EURACHEM document
on "Quantifying Uncertainty in Analytical Measurement" [5] and ISO/TS
21748:2004(E) on “Guidance for the use of repeatability, reproducibility and
trueness estimates in measurement uncertainty estimation”. [10]. Simplified
methods adopted elsewhere are also considered. Guidance is also given on the
expression and reporting of uncertainty values.
1.5 However, it must be noted that although the concept of uncertainty itself is well
accepted, there are different opinions among many learned establishments on how
it should be estimated and, to a lesser extent, how it should be referred to. Hence, it
is anticipated that this Guide will require constant reviewing and updating, in order to
keep up with the most current methodology.
1.6 The appendices accompanying this document are several detailed examples of
uncertainty evaluation processes taken from different areas of chemical analysis.
These examples are intended to illustrate the application of the procedures
described in this Guide.
2.1 The word "uncertainty" means doubt, and thus in its broadest sense "uncertainty of
measurement" means doubt about the validity of the result of a measurement.
2.3 This definition is also consistent with other concepts of uncertainty of measurement,
such as:
- a measure of the possible error in the estimated value of the measurand as
provided by the result of a measurement;
- an estimate characterizing the range of values within which the true value of
a measurand lies.
2.4 When uncertainty is evaluated and reported in a specified way, it indicates the level
of confidence that the value actually lies within the range defined by the uncertainty
interval.
3.1 There is a growing awareness that analytical data for use in any decision process
must be technically sound and defensible. Limits of uncertainty are required which
need to be supported by suitable documentary evidence in the form of statistical
control as for some kind of ‘quality assurance’. When a measurement process is
demonstrated by such statistical control, the accuracy of the process can be implied
to characterize the accuracy of all data produced by it.
3.2 It is a recognized fact that any chemical analysis is subject to imperfections. Such
imperfection gives rise to an error in the final test result. Some of these are due to
random effects, typically due to unpredictable variations of influence quantities, such
as fluctuations in temperature, humidity or variability in the performance of the
analyst. Other imperfections are due to the practical limits to which correction can
be made for systematic effects, such as offset of a measuring instrument, drift in its
characteristics between calibrations, personal bias in reading an analogue scale or
the uncertainty of the value of a reference standard.
3.3 Every time a measurement is taken under essentially the same conditions. Random
effects give rise to random errors from various sources and this affects the
measured value. Repeated measurements will show variation and a scatter of test
results on both sides of the average value. Statisticians say that random errors
affect the precision, or reproducibility. A number of sources may contribute to this
variability, and their influence may be changing continually. They cannot be
completely eliminated but can be reduced by increasing the number of replicated
analysis.
3.4 Systematic errors emanate from systematic effects. They cause all the results to be
in error in the same sense, i.e. either producing consistently higher or lower results
than the true value. They remain unchanged when a test is repeated under the
same conditions. These effects also cannot be eliminated but may be reduced or
corrected with a correction factor if a systematic effect is recognized. In fact,
systematic errors must be first dealt with before estimating any uncertainty in a
chemical analysis.
3.7 By quoting measurement uncertainty, the laboratory operator reflects well on the
technical competence of his laboratory staff performing the analysis and helps to
communicate the limitations of test results to his customer.
c) Incomplete definition of the measurand (e.g. failing to specify the exact form
of the analyte being determined, such as Cr3+ and Cr6+ );
d) Imperfect realization of the definition of the test method. Even when the test
conditions are defined clearly, it may not be possible to produce these
conditions in a laboratory;
5.1 The ISO Guide 98, ISO/TS 21748:2004 and the EURACHEM document have all
adopted the approach of grouping uncertainty components into two categories
based on their method of evaluation, i.e. Type A and Type B evaluation methods.
5.2 This categorization, based on the method of evaluation rather than on the
components themselves, applies to uncertainty and is not substitutes for the words
"random" and "systematic". It avoids certain ambiguities - a random component of
uncertainty in one measurement may become a systematic component in another
measurement that has, as its input, the result of the first measurement. For
example, the overall uncertainty quoted on a certificate of calibration of an
instrument will include the component due to random effects, but, when this overall
value is subsequently used as the contribution in the evaluation of the uncertainty in
a test using that instrument, the contribution would be regarded as systematic.
5.3 Type A evaluation of uncertainty is based on any valid statistical method in analysis
of a series of repeated observations. The statistical estimated standard uncertainty
is called, for convenience, a Type A standard uncertainty.
5.4 Component of Type A evaluation of standard uncertainty arises from random effect.
The Gaussian or Normal Law of Error forms the basis of the analytical study of
random effects. (See Appendix B)
5.5 It is a fact that the mean of a sample of measurement provides us with an estimate
of the true value, µ of the quantity we are trying to measure. Since, however, the
individual measurements are distributed about the true value with a spread which
depends on the precision; it is most unlikely that the mean of the sample is exactly
equal to the true value of the population.
5.6 For this reason, it is more useful to give a range of values within which we are
almost certain the true value lies. The width of the range depends on two factors.
The first is the precision of the individual measurements, which in turn depends on
the variance of the population. The second is the number of replicates made in the
sample. The very fact that we repeat measurements implies that we have more
confidence in the mean of several values than in a single one. Most people will feel
that the more measurements we make, the more reliable our estimate of µ, the true
value.
x = xi / n … (1)
5.8 The experimental standard deviation s is used to estimate the distribution of x as:
5.9 The experimental standard deviation of mean, or standard error of the mean
(s.e.m.), x, or a distribution of sample means has an exact mathematical
relationship between it and the standard deviation, , of the distribution of the
individual measurements, which is independent of the way in which they are
distributed. If N is the sample size, this relationship is:
s.e.m. x = / N … (4)
5.10 From the equation (4) above, it is noted that the larger N is, the smaller the spread
of the sample means about µ. This universally used term, the standard error of the
mean, might mislead us into thinking that / N gives the difference between x
and µ. This is not so. The / N gives a measure of uncertainty or confidence
involved in the estimating µ from x .
5.11 On the other hand, Type B evaluation is by means other than used for Type A such
as:
- from data provided in calibration certificates and other reports;
- from previous measurement data;
- from experience with, or general knowledge of the behaviour of the
instruments;
- from manufacturers’ specifications;
- from all other relevant information.
5.12 When we are considering Type B uncertainty, we have to convert the quoted
uncertainty to a standard uncertainty expressed as standard deviation. We can
convert a quoted uncertainty that is a stated multiple of an estimate standard
deviation to a standard uncertainty by dividing the quoted uncertainty by the
multiplier.
Example:
A calibration report for reference weights states that the measurement uncertainty of
a 1-gm weight is 0.1 mg at 2 standard deviations. The standard uncertainty is
therefore 0.1 mg divided by 2 which gives 0.05 mg.
5.14 It may be stressed that those uncertainty components quantified by means other
than repeated analysis are also expressed as standard deviations, although they
may not always be characterised by the normal distribution. For example, it may be
possible only to estimate that the value of a quantity lies within bounds (upper or
lower limits) such that there is an equal probability of it lying anywhere within those
bounds. This is known as a rectangular distribution. There are simple mathematical
expressions to evaluate the standard deviation for this and a number of other
distributions encountered in measurement.
5.15 The components, evaluated by either Type A or Type B methods, are combined to
produce an overall value of uncertainty known as the combined standard
uncertainty. An expanded uncertainty is usually required to meet the needs of
industrial, commercial, health and safety, and other applications. It is obtained by
multiplying the combined standard uncertainty by a coverage factor, k. The k value
can be 2 for a 95% confidence level and 3 for a 99.7% confidence level. The
expanded uncertainty defines an interval about the result of a measurement that
may be expected to encompass a large fraction of the distribution of values that
could reasonably be attributed to the measurand.
6.1 Before the discussion on the methods for estimating uncertainty, it is helpful to first
of all break down the analytical method into a set of generic steps in order to identify
the possible sources of uncertainty:
a. Sampling
b. Sample preparation
c. Use of certified reference materials to the measuring system
d. Calibration of instrument
e. Analysis for data acquisition
f. Data processing
g. Presentation of results
h. Interpretation of results
6.2 Each of these steps can be further broken down by contributions to the uncertainty
for each. The following list, though not exhaustive, demonstrates the various factors
that need to be considered when determining the sources of measurement
uncertainty.
6.2.1 Sampling
- The physical state of the population (bulk) for sampling (gas, liquid or solid);
- Is the population (bulk) static or flowing?
- Does the population consist of discrete units?
- How homogeneous is the population (bulk)?
- Any temperature and pressure effects;
- Physical and chemical stability of the sample.
6.2.5 Analysis
- Analyst personal effects for systematic errors (e.g. colour blindness)
- Avoidance of contamination and cross-contamination
- Reagent purity
- Instrument parameter settings, e.g. GC conditions
7.1 Uncertainty estimation is simple in principle. Appendix C shows the flow diagram of
the evaluation process. The following steps summarise the tasks that need to be
performed in order to obtain an estimate of the uncertainty associated with a
measurement.
7.2.3 Hence, a chemical measurement result is obtained at the end of a series of steps in
a procedure. This is a numerical value for the measurand that is dependent upon a
number of intermediate or input quantities. These may be other measurands or
constants (constants also have uncertainties).
7.2.4 In general, the measurand has a relationship to these other quantities which, in
principle can be expressed algebraically as:
x = f(a,b,c,…) … (5)
7.2.5 Such an approach is useful for a theoretical discussion but in practice, except in the
simplest cases, it is rarely utilized. It is more useful to break down the
measurement procedure into a number of blocks. The results of the uncertainty
evaluations on these simple blocks can then be used to obtain the combined
uncertainty.
7.2.6 Therefore, for the purpose of uncertainty calculations, it is advisable to break down
the relationship between the measurand and the input quantities into simple
expressions that conform to one of the standard probability distribution forms as
stated in Appendix B.
a) Sample Uncertainty
The uncertainties in the data due to the sample always need to be
evaluated. The sampling operation from a population can introduce both
systematic and random errors. Calibration errors can cause problems of
the first type, whilst variability of operation such as sieving or extraction are
examples of the latter kind, it may be impossible to quantify the individual
components of sampling variance. However, the overall sampling variance
can be evaluated by taking a number (at least 7) of samples under
conditions where the samples are expected to be essentially identical. The
total variance consists of the sum of that due to the samples and to their
measurement. Thus:
b) Instrument Bias
Systematic errors can occur in an analytical instrument. Many analysts tend
to make false assumption about the accuracy of the instrument. For
example, the monochromators in spectrometers gradually go out of
adjustment, so that errors of several manometers in wavelength settings are
not uncommon, yet many photometric analyses are undertaken without
appropriate checks being made. Very simple devices such as stop-
watches, pH meters and thermometers can all show substantial systematic
errors too.
d) Human Bias
Systematic errors can also arise from human bias. Some chemists suffer
from astigmatism or colour-deficiencies which might introduce errors into
their readings of instruments and other observations. Serious errors can be
made by them in the titration process using colour indicator.
e) Computational Effects
The increasing availability of instruments controlled by micro-processors or
microcomputers has reduced to a minimum the number of operations and
the level of skill required of their operators. In these circumstances the
temptation to regard the instruments’ results as beyond reproach is
overwhelming. However, the uncritical use of computer software can
introduce errors into the reported results as the programmes are subject to
conceptual errors such as coding population instead of sample standard
deviation. There may be error in selecting an inappropriate calibration
model, e.g. using a straight line calibration on a curved response. Early
truncation and rounding off can also lead to inaccuracies in the final test
result.
f) Calibration Uncertainties
Ideally, the calibration process is undertaken to eliminate deviations in the
accuracy of measurements or instruments. However, this cannot be glibly
assumed. In fact, as the limits of measurement are approached, the
uncertainties of calibration may increase in a similar manner and can be the
limiting factor in attainable accuracy.
g) Cross Contamination
In any trace analysis, analysts must be fully aware of the possibility of cross
contamination between samples and contamination from the laboratory
environment as a result of poor working practices. For example, in the
analysis of trace volatile organic compounds by headspace or purge and
trap technique, any solvent extraction process in the nearby vicinity of a gas
chromatograph will certainty affect the final results. Hence, such a risk of
uncertainty must be minimized whenever possible.
7.3.3 It is a good practice to write down all possible sources of uncertainty in a chemical
analysis and then simplify by re-grouping them under more general headings. For
example, instead of writing down temperature, pressure and calibration errors as
sources of uncertainty for every determination of weight by difference, it may be
more sensible to regard all these three factors as parts of a single heading:
‘weighing uncertainty’ which can be evaluated directly.
Technical Guide 2, Mar 08 10
7.4 Step 3 - Quantifying Uncertainty
7.4.1 It is to be aware that not all the components of uncertainty are going to make a
significant impact to the combined uncertainty to be evaluated. Indeed, in practice,
it is likely that only a small number will. Hence, the first step in the quantification of
uncertainties identified is to make a preliminary estimate of the contribution of each
component to the combined uncertainty and to eliminate those which are not
significant. Regroup certain sources of uncertainty with a view of simplification and
evaluate them as a single component.
7.4.2 Calibration process for example, will provide a combined uncertainty associated
with the blocks of the measurement process and a detailed evaluation of each
component within the block is therefore not necessary.
7.4.3 Where uncertainty sources are grouped in this manner, the groups should be
identified and the uncertainty sources included should be checked against the list
generated by Step 2. This provides an auditable record of which contributions to
uncertainty have been included.
7.4.4 Four basic methods can be used to estimate the individual uncertainty component:
7.4.5.2 However, it must be stressed that before any repeated experiments are to be
carried out for this purpose, systematic errors present, in any, which occur in a
definite and known sense, must first of all be dealt with.
7.4.5.3 The analyst must consider in the beginning of a measurement in an experiment the
likely sources of systematic error such as the instrument functions that need
calibrating, and the steps of the analytical procedure where errors are most likely to
occur. He should also consider the checks that he can subsequently make for
systematic errors. The most formidable protection against systematic errors is the
use of standard reference materials and methods. If a non-standard method is to
be used, then it is a good practice to compare the results of the method against
those obtained by another chemically and physically unrelated method or a
standard reference method. If both methods consistently yield results
showing only random differences, it is a reasonable presumption that no
significant systematic errors are present. On the other hand, if systematic
differences do occur, a correction factor for the non-standard method has to be
established after repeated analysis.
7.4.5.4 In most analytical procedure, only a few components of the uncertainty dominate.
Hence, it is realistic to vary these parameters to the fullest practicable extent in the
experiment so that the evaluation of uncertainty is based as much as possible on
observed data.
7.4.6.2 As the reference material will be used for measurement quality assessment, its
property certified must be accurately known. The uncertainty in the certified values
takes into account that due to measurement and any variability (such as non-
homogeneity) between and/or within sample of the material. Thus, definitive
methods are used for establishing the values of the certified properties or they are
measured by two or more independent reliable methods. Hence, the uncertainty of
the values stated is quite minimal.
7.4.6.3 Therefore, measurements on such reference materials provide very good data for
the assessment of uncertainty since they provide information on the combined
effect of many of the potential sources of uncertainty. This method of
measurement uncertainty is recommended if reference material is available.
7.4.6.4 However, there are other sources of uncertainty in such process that have to be
taken into account, such as:
7.4.6.5 To overcome the matrix effects, the analyst can use the most recent inter-
laboratory cross-check samples of similar nature as the reference material.
For sources of uncertainty not adequately covered by existing data, one may
obtain additional information from the literature or standing data (certificates,
equipment specifications, etc.). One may also plan experiments to obtain
the required additional data. Additional experiments may take the form of
Technical Guide 2, Mar 08 12
specific studies of a single contribution to uncertainty, or the usual method
performance studies conducted to ensure representative variation of
important factors.
It may be noted that repeatability data are used as a check on precision, which, in
conjunction with other tests, confirms that a particular laboratory may apply
reproducibility and trueness data in its estimation of uncertainty.
The ISO/TS 21748:2004 lists out the following procedure for evaluating
measurement uncertainty:
b) establish whether the laboratory bias for the measurements is within that
expected on the basis of the data obtained in (a);
e) where the bias and precision are under control, as demonstrated in (b) and
(c), combine the reproducibility estimate (a) with the uncertainty associated
with trueness [(a) and (b)] and the effects of additional influences (d) to
form a combined uncertainty estimate.
It should be stressed however, that one must not doubly estimate a measurement
uncertainty. If the approach of this clause is adopted, the component-by-
component approach taking the uncertainty budgets of individual analytical steps
should be carefully evaluated. As the reproducibility standard deviation taken from
a collaborative study might have been obtained from samples of different matrices,
the matrix effect would have been taken care of but one may have to investigate
separately to ensure the laboratory bias is under control, such as via the recovery
study.
7.4.8.2 Judgement of this type is not based on immediate experimental results, but rather
on experience with, or general knowledge of the behaviour and property of relevant
materials and instruments. It is quite a subjective probability, an expression which
can be used synonymously with ‘degree of belief’, ‘intuitive probability’ and
‘credibility’. Such degree of belief is not based on a snap judgement, but on a well
considered mature professional judgement of probability through expert knowledge
by earlier experiments and observations. It constitutes typical Type B evaluation as
it does not rely on replicated experiments performed just for a specific evaluation of
uncertainty.
7.4.8.3 For the purpose of estimating combined uncertainty, two features of degree of
belief estimations are essential:
Example
From the standard table giving z values and areas under the Standard
Normal Probability Distribution, a figure of z=1.96 is found to give 95.0%
area under curve. Using this figure gives a standard uncertainty of
(0.2/1.96) = 0.1.
7.4.9.3 Where an estimate is made upon judgement, it may be possible to estimate the
component directly as a standard deviation. If this is not possible, an estimate
should be made of the maximum value which could reasonably occur in practice.
7.5.2 The standard uncertainty of y, where y is the estimate of the measurand Y and
thus the result of the measurement, is obtained by appropriately combining the
standard uncertainties of the input estimates a,b,c,…. This combined standard
uncertainty of the estimate of y is denoted by uc(y).
7.5.3 The combined standard uncertainty, uc(y) is the positive square root of the
2
combined variance uc (y) which is given by:
uc2(y) = {[ f/ a]2u2(a) + [ f/ b]2u2(b) + [ f/ c]2u2(c) + … ... (8)
Technical Guide 2, Mar 08 15
7.5.4 For practical purposes, the following simple rules for combining standard deviations
are shown below:
2 2 2
u(y) = k [u (a) u (b) u (c) ......] … (9)
2 2 2
u(y)/y = k [u(a)/a] [u(b)/b] [u(c)/c] ...... … (10)
7.5.5 The final stage is to multiply the combined standard uncertainty by the chosen
coverage factor in order to obtain an expanded uncertainty. The coverage factor is
chosen after considering a number of issues like the level of confidence required
and any knowledge of underlying distributions. For most purposes, a coverage
factor k of 2 is normally chosen which gives a confidence level of approximately
95%.
7.5.6 The expanded uncertainty is required to provide an interval which may be expected
to encompass a large fraction of the distribution of values which could reasonably
be attributed to the measurand.
In such a case, the laboratory should carry out a series of tests to evaluate the
impact of such factors on the results. One common example is the temperature
effect on water volume expansion. Nowadays, analysts often use water expansion
coefficient of 2.1 x 10-4 per °C per ml to estimate measurement uncertainty of
volume due to temperature change. Such water expansion coefficient is in fact a
sensitivity coefficient. It was obtained by changing the temperature and accurately
determining the resultant volume. From the volume changes, one can work out the
volume change per °C per unit volume. Similarly, oven temperature effect and
moisture content can be determined in such a manner.
a) a description of the methods used to calculate the result and its uncertainty;
b) the values and sources of all corrections and constants used in both the
calculation and the uncertainty analysis;
The data and analysis should be presented in such a manner that its important
steps can be readily followed and if necessary repeated.
Note that the texts within square brackets [ ] may be omitted or abbreviated in a
suitable manner.
Example
Total Oil Content : 9.80 ± 0.15% w/w*
8.3.3 Although a coverage factor of 2 is commonly used for 95% confidence reporting,
coverage factors of either k =1 or k = 3 may be considered in some cases. These
correspond to confidence levels of 68% and more than 99%, respectively.
9.1 It is important that one should not doubly count an uncertainty component. When
the GUM approach is adopted where the standard uncertainty of each component is
considered fully, one should not introduce the repeatability and reproducibility of the
test method as the other uncertainty components because repeatability and
reproducibility represents the total performance of the test method in terms of its
precision and accuracy, respectively.
9.3 The GUM approach does not take the possible result bias into consideration but in
chemical analysis, there is always a possibility of systematic error which could then
be minimized or eliminated. One has to either estimate a correction factor to adjust
the test result back into its true value or estimate the bias uncertainty of reporting
such biased result.
10.1 Introduction
10.1.1 This Guide adopts the “top-down” or “global” approach to MU, which is based on
the standard deviation of reproducibility of the final result of the measurement
process. The same approach has been endorsed for a more general use by
ISO/TS 21748 elaborated by ISO/TC 69, Application of Statistical Methods, SC
6, Measurement Methods and Results [11], and detailed in the ISO/TS
19036, ”Microbiology of Food and Animal feeding stuffs – Guidelines for the
Estimation of Measurement Uncertainty for Quantitative Determinations” [12]. The
latter document clarifies that the step-by step approach and the global approach
are not mutually exclusive, since all the MU components can be considered to be
included in the overall performance of the analytical process, which can be
characterized by the observable precision and bias.
10.1.2 The current consensus has been that the “step-by-step” approach does not apply
satisfactorily in the case of microbiological analysis, where it is difficult to build a
really comprehensive model of measurement process. It appears difficult to
quantify accurately the MU contribution of each individual step of the
microbiological measurement process, where (1) the analyte is a living organism
present in a natural sample that can be in variable physiological state in their
natural environment, e.g. in various stage of growth or in injured condition on
exposure to the adverse environmental conditions or manufacture processes, (2)
the target organism includes different strains, different species or different
genera, and (3) there are no truly certified reference preparations of micro-
10.2 Scope
10.2.1 This Guide provides guidance for the estimation and expression of MU
associated with quantitative microbiological methods, in which the quantitative
estimate is based on counting of particles on the basis of growth (multiplication)
into colonies. These methods are commonly known as the heterotrophic plate
count, total aerobic microbial count, spiral plate count (instrument method), and
colony counts of specific target organisms on selective media, e.g. faecal coliform
count, and coagulase-positive Staphylococcus count
10.3.2 The following sources of uncertainty have been shown to influence the precision
and hence measurement uncertainty of microbiological results:
10.3.2.1 Sampling
of sample
a) Nature
Homogeneity of the sample
Background microflora and their concentration over the target
organism to be counted
d) 7UDQVSRUWDWLRQWLPHDQGWHPSHUDWXUHRIVDPSOH
b) Definition of measurand
10.3.2.4 Procedure
Analytical
Sample homogenization/mixing, e.g. using Stomacher, blender or
turbo mixer.
Sub-sampling
Preparing and dispensing dilutions
Inoculation procedure, e.g. Filtration technique, pour-plate, spread-
plate and spiral plate techniques
Incubation conditions
Reading, interpreting and reporting results
Microbial concentration in each culture plate
10.3.2.5
Equipment
Precision and accuracy
Maintenance
Calibration
Repair
10.3.2.6
Analysts
Training
Validating & Maintaining Competency
10.4.2 Identify individual components of uncertainty and demonstrate that they are under
control, for example, regular checking of performance of culture media,
incubators, weighing balance, pipettors and other instrument, and within-analyst
repeatability.
10.4.4 Perform analysis of at least 15 samples that are set up on different days in
duplicate pairs, different analysts on different days, using different equipment
(e.g. pipettors, incubators, if more than one pipettor or incubator is used for the
same test) on the different days and possibly using different batches of reagents
and media on different days.
Duplicate pair analysis refers to the analysis of the same sample two times, each
time using the same procedure by the same analyst on the same day within a
short period of time over which the level of the microorganisms remains stable.
10.4.5 The data should be collected over an extended period of time, e.g. 1 year. The
samples should consist of low, medium and high concentrations of
microorganisms that normally encountered in the natural samples. The
recommended counting range of colonies per plate stipulated in the standard
methods can be used as a starting point.
10.4.6 Natural samples should be used as far as possible, since they enable a more
realistic estimation of MU. If spiking is required, spikes should be designed to
mimic natural contamination as far as possible, e.g. by use of organisms
harvested and concentrated from fluid sample by centrifugation. When this is not
feasible, reference organisms may be used as spikes.
n
RSDR = [ ( (log ai-log bi) / xi)2 / 2n] … (11)
i=1
where
(log ai-log bi)/ xi = the relative difference between the duplicate logarithmic
results
i =1,2,… .n
n = number of duplicate pairs in the analysis
In this case the relative difference in each pair is calculated. This is done by
dividing the difference of each pair with the mean value of the pair. This
difference is then squared. These are summed. The sum is divided by 2 times
the number of duplicate determinations. The square root is taken for that value to
give the Relative Standard Deviation of Reproducibility (RSDR).
10.4.8 The combined uncertainty associated with the procedure is the value of RSDR :
u = RSDR
U = 0.04 x 100 = 4%
To obtain the MU for the count, c, then use the following equation,
The range for count/g after round up would be 2.0 x 104 to 4.5 x 104
During any intermediate stages in the calculations, e.g. when transforming counts
to log10 values, calculating the mean, etc, try and keep figures as accurate as
possible and only round the final results to the desired precision. For colony
counts, not more than two significant figures shall be used for reporting the
values of the result and the uncertainty interval.
The calculation of the MU from the above equation produces a range of common
logarithms of the count. This range is that which could reasonably be expected if
similar samples were to be tested by same group of operators within this
laboratory, and the counts of these samples fall within the range or
concentrations of counts used in the RSDR estimation.
Using the above example : Colony forming units /g : 3.0 x 104 with confidence
interval of 2.0 x 104 to 4.5 x 104
10.4.10 This approach assumes RSDR is constant over the recommended counting
range of the method. Where sufficient data is available, this assumption should
be verified for subsets of data of duplicate pairs obtained by the method by
comparing RSDR values for low, medium and high concentration ranges.
10.5 The Standard Grubbs Test for Identification of Outliers of Duplicate Pairs
10.5.1 Examine the dataset of duplicate pair analysis, such as the cfu count of duplicate
plates and the relative difference of counts/mean for suspected outlier. The
Standard Grubbs Test for identification of outlier of duplicate pairs can be used to
determine whether suspected outlier can be reasonably removed, at a selected
risk of false rejection. The Grubbs test calculates how much a suspected outlier
differs from the population mean, measured in units of standard deviation.
T = I RD I / ( 2 x RSDR) … (12)
where
RD = Relative difference for each duplicate pair, as a decimal fraction
[RD = (ai-bi) / x ]
10.5.3 Compare the calculated T value to the appropriate critical value in the following
Table based on the number of data points in the set and the risk that can be
tolerated for false rejection. A 5% risk of false rejection is recommended. In the
Table, the “Number of Data Points” column refers to the number of duplicate pairs
used in the calculation of RSDR.
If outliners are excluded, re-calculate the RSDR with the outliners removed.
Example:
Pour Plate Counting using Standard Method Estimation Dairy Product(SMEDP):
U =kxu
= 2 x 19.8%
= 39.6%
Note: When using the standard method, the laboratory is required to demonstrate
their ability to meet the established performance requirements of the
standard method, a pre-requisite for making use of the established
expanded uncertainty of the standard method.
10.7.2 A 2 (chi-square test) can be used to find out whether a microbiological result has
exceeded the stipulated microbial limit:
2 = (C-L)2 / L
where C = colony count, L = Limit value
10.8.2 For the purposes of SAC-SINGLAS’s Policy, the McCrady’s table can be used as
estimates of uncertainty for a test, provided the laboratory has reviewed the
resulting data and identified any unusual combinations of results.
10.8.3 Any unusual combinations of positive and negative tubes in excess of 1% of all
MPN results are to be treated as non-conforming to the McCrady’s table. Root
causes should be identified and corrected.
10.10.2 The expanded uncertainty determined from data over the entire counting range of
colonies per filter or plate may overestimate or underestimate uncertainty
depending upon whether the data is weighted to high or low counts. Canadian
Association for Environmental Analytical Laboratories (CAEAL) Uncertainty Policy
[14] recommended that data be separated into ranges (as indicated below) and,
the combined uncertainty (U) determined for each range.
The following colony forming unit (cfu) ranges are suggested for membrane
filtration techniques for water samples:
-19 cfu/Filter
20-80 cfu/Filter
-100 cfu/Filter
The following cfu ranges are suggested for plating (e.g. spread plate) procedures
for water samples:
1-29 cfu/Plate
-99 cfu/Plate
-300 cfu/Plate
The following definitions of statistical terms are quoted from the following documents:
ISO/IEC : Guide 2
A.1 Accuracy
The closeness of agreement between a test result and a true value of the measurand.
A.2 Analyte
The specific component measured in a chemical analysis.
A.4 Bias
The difference between the expectation of the test results and an accepted reference
value.
Note 1: Bias is the total systematic error as contrasted to random error. There may
be one or more systematic error components contributing to the bias. A
larger systematic difference from the accepted reference value is reflected
by a larger bias value.
A.5 Calibration
Comparison of a measurement standard or instrument with another standard or
instrument to report or eliminate, by adjustment, any variation or deviation in the
accuracy of the item being compared.
s
CV =
x
In a control chart, the limits below which (upper limit) or above which (lower limit) or
the limits outside which the statistic under consideration lies when action should be
taken.
The warning limits are always within the action limits and are between the upper
and/or lower limits and the central line. When the value of the statistic computed
from a sample is outside the warning limits but inside the action limits, increased
supervision of the process is generally necessary and rules may be made for action
in particular processes. In other words, at the warning limits, attention is called to the
possibility of out-of-control conditions, but further action is not necessarily required.
A.16 Deviation
Difference between a value and its reference value.
A.17 Drift
Moderate changes in the measurement characteristics of a measuring instrument.
Note 2: Like true value, systematic error and its causes cannot be known.
XL = Xblank + k sblank
Where Xblank is the mean of the blank measures and sblank, the standard
deviation of the blank measures, and k, a numerical factor chosen
according to the confidence level desired.
A.25 Measurand
Particular quantity subject to measurement.
A.26 Measurement
Set of operations having the object of determining a value of a quantity.
A.29 Metrology
Scientific execution of measurement.
A.31 Non-conformity
The non-fulfillment of a specified requirement.
A.32 Outlier
A value which appears to deviate markedly from that for other members of the
sample in which it occurs.
A.33 Population
A generic term denoting any finite or infinite collection of individual things, objects or
events. It is the totality of items under consideration.
A.34 Precision
The closeness of agreement between independent test results obtained under
stipulated conditions.
A.36 Probability
The likelihood of the occurrence of any particular form of an event, estimated as the
ratio of the number of ways or times that the event may occur in that form, to the total
number of ways that it could occur in any form.
A.41 Recovery
The fraction of analyte added to a test sample (fortified or spiked sample) prior to
analysis, the unfortified and fortified samples, the percentage recovery (%R) is
calculated as follows:
CF - CU
%R = -------------- x 100
CA
A.44 Repeatability
Precision under repeatability conditions, i.e. conditions where independent test
results are obtained with the same method on identical test items in the same
laboratory by the same operator using the same equipment within short intervals of
time.
A.45 Replicate
A counterpart of another, usually referring to an analytical sample or a measurement.
It is the general case for which duplicate, consisting of two samples or
measurements, is the special case.
A.46 Reproducibility
Precision under reproducibility conditions, i.e. conditions where test results are
obtained with the same method on identical test items in different laboratories with
different operators using different equipment.
Note 1: When the term “result of a measurement” is used, it should be made clear
whether it refers to:
- The indication
- The uncorrected result
- The corrected result
and whether several values are averaged.
A.48 Sample
A portion of a population or lot. It may consist of an individual or groups of
individuals; it may also refer to objects, materials, or to measurements conceived to
be part of a larger group (population) that could have been considered.
A.49 Sampling
The process of drawing or constituting a sample.
A.51 Specification
Document that prescribes the requirements with which the product, process or
service has to conform. It is desirable that the requirements be expressed
numerically in terms of appropriate units together with their limits.
s = [ (xi- x )2/(n-1)]
= [ (xi-µ)2/N]
The terms “standard error” or “standard deviation of the mean” have also been used
to describe the same quantity.
For a normally distributed population with mean and standard deviation , the
standard deviation
X
of the sample mean x if N samples taken from that
population are given by:
X
= / N
A.57 Sub-sample
It is a portion taken from a sample. A laboratory sample may be a sub-sample of a
gross sample; similarly, a test portion may be a sub-sample of a laboratory sample.
A.62 Traceability
The property of the result of a measurement or the value of a standard whereby it
can be related to stated references, usually international or national standards,
through an unbroken chain of comparisons; all having a stated uncertainty.
Note 1: A Type A standard uncertainty is obtained by taking the square root of the
statistically evaluated variance.
Note 1: The parameter may be, for example, a standard deviation (or a given multiple
of it), or the width of a confidence level.
Note 2: To associate a specific level of confidence with the interval defined by the
expanded uncertainty requires explicit or implicit assumptions regarding
the probability distribution characterised by the measurement result and its
combined standard uncertainty. The confidence level that may be
attributed to this interval can be known only to the extent to which such
assumptions can be justified.
U = k uc
A.70 Variance
A measure of dispersion, which is the sum of squared deviations of observations
from the average divided by one less than the number of observations, i.e. variance
is the square of standard deviations.
2
Note 1: The symbols for sample and population variances are s and 2,
respectively.
A.71
-Risk
The chance of rejecting the null hypothesis when the null hypothesis is true, i.e.
concluding that there is a difference between treatments when no difference actually
exists, as stated in the null hypothesis.
DISTRIBUTION FUNCTIONS
B.1 The followings are some common probability distribution functions that can be
used to calculate a standard uncertainty. It may be noted that a probability
distribution gives the probability for each of the values of a random variable. The
choice of an appropriate distribution function depends on the knowledge of the
probability distribution of the uncertainty. The standard uncertainty is then
obtained by dividing the quoted uncertainty by a factor, which depends on the
probability distribution.
Commonly Used Confidence Coefficients & Confidence Intervals for a Large Sample
START
1
No
Yes
Standard
Are comparison Test Method
Yes study data No has stated
available?
4 repeatability/ 8
reproducibilit
y information
Has No
comparison No Yes
study covered 5
all aspects of
test process? Convert method 9
repeatability/
reproducibility data into
standard uncertainty
Yes
Use Step-by-Step 10
method to work out Combine the 11
standard uncertainty standard
Is lab’s uncertainties from
within-lab for those portions of
No test process not these sources to
repeatability 6 obtain combined
comparable? covered by Standard
method or comparison standard uncertainty
study
Yes
Convert 7
comparison
study’s
repeatability /
reproducibility
data into standard
uncertainty
Step 2 Is Test Laboratory following Standard Test Method? If yes, proceed to Step 8. If
no, proceed to Step 3.
Step 3 Has the test method been validated? If yes, proceed to Step 4. If no, Test
Laboratory should validate the test method first before it attempts to estimate
measurement uncertainty, proceed back to Step 1.
Step 4 Are comparison study data available for the test method? If yes, proceed to Step 5.
If no, proceed to Step 10.
Step 5 Has the comparison study covered all aspects of test process? If yes, proceed to
Step 6. If no, proceed to Step 10. If a comparison study covers only parts of the
whole test process, for those portions covered, Test Laboratory may still proceed
to Step 6. For the remaining portions not covered, the Laboratory need to proceed
to Step 10. However, the Laboratory may proceed to Step 10 for the whole test
process without taking into consideration of comparison study data.
Step 8 Have the Standard Test Method stated repeatability / reproducibility information? If
yes, proceed to Step 9. If no, proceed to Step 4. If the Standard Test Method
covers only parts of the whole test process, for those portions covered, Test
Laboratory may still proceed to Step 9. For the remaining portions not covered, the
Laboratory need to proceed to Step 4. However, the Laboratory may proceed to
Step 4 for the whole test process without taking into consideration of the data from
the Standard Test Method.
Step 9 Test Laboratory converts the Standard Test Method’s repeatability / reproducibility
data into standard uncertainty (see Working Example F5). Proceed to Step 11.
Step 11 Test Laboratory then combines the standard uncertainties from all aspects
covered under Steps 7, 9 and 10 to obtain the combined standard uncertainty.
Step 12 Test Laboratory converts the combined standard uncertainty into the expanded
uncertainty at appropriately 95% confidence level using appropriate Student t-
factor.
Step 13 Test Laboratory shall then document the estimated measurement uncertainty
including the above calculations together with supporting data files to ensure data
traceability for future reference.
Step 14 End.
1. Introduction
1.2 The general approach used in this Technical Specification requires that
- The laboratory confirms that its implementation of the test method is consistent
with the established performance of the test method by checking its own bias
and precision. This confirms that the published data are applicable to the
results obtained by the laboratory;
- Any influences on the measurement results that were not adequately covered
by the collaborative study are identified and the variance associated with the
results that could arise from these effects be quantified.
1.4 The ISO/TS 21748 assumes that recognised, non-negligible systematic effects are
corrected, either by applying a numerical correction as part of the method of
measurement, or by investigation and removal of the cause of the effect.
2. General Principles
2.1.2 The ISO/TS 21748 requires that process performance figures derived from
method-performance studies are relevant to all individual measurement results
produced by the process. It will be seen that this condition requires supporting
evidence in the form of appropriate quality control and assurance data for the
measurement process.
2.2.1 The application of the principles of the ISO/TS 21748 is based on two principles
- Second, effects not observed within the context of the collaborative study must
be demonstrably negligible or explicitly allowed for.
The following generic worked examples are intended to show how the principles of this
Guide can be applied in specific analytical processes. The readers are encouraged to
prepare their own generic worked examples to suit their own needs and make their decision
on which causes of errors can be ignored due to insignificant contribution to the overall
uncertainty to be measured.
Readers will appreciate the convenience of using computer spreadsheet in this aspect and
are strongly recommended to store all information of each contributing component and
cross-link all these components in one spreadsheet for each test method.
E.1.1 Purpose
Weight WCu
The supplier quotes the purity of the copper wire in its certificate of analysis as 99.99
± 0.01% without mentioning its degree of confidence.
As there is no idea of the confidence limit of this purity, we take the quoted
uncertainty as the rectangular distribution, so the standard uncertainty u(PCu) is:
0.0001
--------- = 0.000058
3
The external calibration of the balance used states that the difference
from the actual weight on the scale pan and the reading of the scale is within ± 0.05
mg with a 95% confidence.
0.05
------- or 0.026 mg
1.96
NOTE: This component uncertainty has to be taken into account twice because of
two weighings involved, one before adding the copper metal and one after.
E.1.3.2.2 Repeatability
E.1.3.2.3 Sensitivity
Sensitivity of the balance can be neglected because the weight by difference is done
on the same balance over a very narrow range.
2 2
u(mCu) = 2 (0.026) (0.06) = 0.07 mg
2 2
= 0.000058 0.00014 = 0.00015
For the copper weight of 500.7 mg, the report of uncertainty is therefore,
Note: that the uncertainty contribution of purity of the copper metal is quite small and
can be neglected.
E.2.1 Purpose
E.2.1.1 To prepare an acid digested copper nitrate standard solution from 500.7 mg copper
wire in 500 mL volumetric flask.
U(W)/W
E.2.3.1 Weighing uncertainty has been previously established at 500.7 mg ± 0.15 mg with a
coverage factor of 2.
The manufacturer states that for the flask of 500-mL, the error is ± 0.15 mL at a
temperature of 20oC. As it has been given without any confidence level stated, we
may assume a triangular distribution because the actual volume is more likely to be
at the centre rather than at the extremes of the range.
According to the manufacturer, the flask has been calibrated at 20oC whereas the
laboratory temperature varies between the limits of ± 4oC. The uncertainty from
this effect can be calculated from the estimate of temperature range and the
coefficient of volume expansion. As the volume expansion of liquid is larger than
that of the flask, only the liquid expansion needs to be considered here. Take the
coefficient of expansion of water as 0.00021per oC.
2 2 2
u(V) = 0.06 0.04 0.24 = 0.25
2 2
u(Conc)/Conc = [u(W)/W ] [u(V)/V ]
2 2
= [0.07 / 500.7] [0.25 / 500]
= 0.00052
E.2.6 Remarks:
E.2.6.1 As it can be seen from the above, contribution of uncertainty in weighing is much
smaller than that in volume preparation.
E.3.1 Purpose
IUPAC has published a list of elements with their individual atomic weight and
associated uncertainty in its journal Pure Appl. Chem., Vol 69, pp. 2471-2473 (1997).
A list of common elements is quoted below for easy reference:
A complete list of all elements and their uncertainties can be found on website:
http://www.chem.qmw.ac.uk/iupac/AtWt/
E.3.3.1 Atomic weights and listed uncertainties (from IUPAC tables) for the constituent
elements of KMnO4 are as follows:
= 158.0339 g.mol-1
2 2 2
u(MWKMnO4) = 0.000058 0.0000052 ( 4x0.00017 )
= 0.0007 g.mol-1
E.3.4 The elemental contribution to KMnO4 is simply the sum of the single atom
contributions. Hence, combined uncertainty would be calculated as a square root of
the sum of squares of each contributing atom.
In some cases, analytical methods require such linear relationship with forced-zero,
i.e. intercept a =0. In these cases, linear relationship is y = bx and xobs = yobs/b.
There are four main sources of uncertainty to consider when estimating uncertainty
of xobs:
Of these four sources, the most significant one is (a). Method for estimating (a)
introduced below is through variance of residuals, S. S can be calculated from
S2 = (yi-yc)2/(n-2)
whereby
For this example, 3 levels of calibration standards are used and their responses are:
Concentration, xi Response, yi
5 125
50 1,197
200 4,754
b=
∑ x y − nx y
i i
a= y − b x
∑ x − nx 2 2
i
In this example,
x y xy x2
5 125 625 25
50 1197 59850 2500
200 4754 950800 40000
Sum 255 6076 1011275 42525
Average 85 2025.333
Therefore,
b =
∑ x y − nx y
i i
=
1011275 − 3 × 85 × 2025.333
= 23.732
∑ x − nx 42525 − 3 × 85 2
2 2
i
With this equation, one can work out calculated responses yc with known x and their
corresponding square of difference (y-yc)2:
x Y Calculated yc (y-yc)2
5 125 126.76259 3.10672
50 1197 1194.7086 5.25036
200 4754 4754.5288 0.27961
Thus,
S2 = (yi-yc)2/(n-2)
= (3.10672+5.25036+0.27961)/(3-2)
= 8.63669.
var(x) = S2/b2
= 8.63669/23.73213432
= 0.0153346
u(xobs,y) = var(x)
= 0.0153346
= 0.124
E.5.1 Purpose
E.5.2 The following steps are taken to appraise the measurement uncertainty concerned.
The chemist uses the GC-MS technique to analyse biphenyl impurity in benzene.
The standard used for calibration is a 50 µg/mL standard solution and a blank
solution (i.e. 0 µg/mL).
The concentration (µg/mL) of biphenyl in benzene can be calculated using two-point
calibration method (bracketing method):
where,
Aspl Area response of GC-MS for biphenyl in the sample
A50 Area response of GC-MS for biphenyl in the standard
A0 Area response of GC-MS for biphenyl in the blank
C50 Concentration of biphenyl in the standard solution which has a nominated
concentration of 50 µg/mL biphenyl
precision precision
Cspl
C50:
The standard solution is prepared from biphenyl solid, first by weighing and then by
dissolution and dilution in benzene.
The purity of biphenyl is stated to be more than 99.0% by the supplier. Thus, the
purity of the raw material is calculated to be 99.5% associated with a standard
uncertainty (100%-99%) / 12 = 0.289%.
The solid biphenyl is then dissolved and diluted to 1,000 mL using volumetric flask.
The specification of the nominal 1L volumetric flask used states the accuracy of
Technical Guide 2, Mar 08 49
1000.22 ± 0.20 mL. In this case, a triangular distribution is used as past QC checks
has shown that centre distribution is more likely than those near the bounds. Thus,
the standard uncertainty from glassware certification is 0.20 / 6 = 0.0816 mL.
Repeated filling and weighing has shown a standard uncertainty of 0.15 mL.
For temperature variation effect on the volume of benzene used for dissolution and
dilution, as there is no available data on expansion coefficient of benzene, we
assume that its expansion is about twice as much as water at ambient temperature
(which has an expansion coefficient of 2.1x10-4 per C). From our experience, we
know that this estimation should be sufficient.
2 2 2
(0.0816 0.15 0.840 ) = 0.857 mL.
m P V
Value 0.052 99.5% 1000.22
Uncertainty 0.000206 0.289% 0.857
Linearity:
1.0/ 3 = 0.577.
First, the standard uncertainties due to C50, A0, A50 and Aspl are combined by
spreadsheet method as shown in the next page to give a concentration in the sample
as 34.836 µg/mL with a standard uncertainty of 0.266 µg/mL.
Thus, the total combined standard uncertainty is (0.2222 0.5772 ) = 0.618 µg/mL.
As C50=1000000 x mP/V,
thus,
m P V A0 A50 Aspl
Value 0.052 99.5% 1000.22 0.9 394.4 265
Uncertainty 0.000206 0.289% 0.857 0.277 0.833 0.869
E.6.1 In general, test recovery is defined as: “Proportion of the amount of analyte, present
in or added to the analytical portion of the test material, which is then extracted and
presented for measurement.” Such recovery studies form an essential component
of the validation and use of all analytical methods to check their accuracies.
E.6.2 Recovery data R is obtained as the ratio of the concentration of analyte found by the
method to that stated to be present (known or true value) and can be used to
determine the bias present, if any, in that particular test method. If the bias does
exist (i.e. the test results obtained are consistently higher or lower than the true
value), an investigation must be made to find out the cause of such systematic error
and minimize it, if possible. If not, the test result must be adjusted by a correction
factor.
E.6.3 To get a recovery data, one has to add a known amount of analyte to a matrix and
the whole matrix is then subject to a normal analysis. The amount recovered minus
the original amount present should indicate the recovery factor.
E.6.4 In a perfect situation, R would be exactly unity (1) but in reality, circumstances such
as imperfect extraction often give observations that differ from the ideal.
E.6.5 Hence, we must take note of the sources of uncertainty in recovery estimation.
Some of them are:
E.6.6 We can test the recovery for any significant departure from unity by the t-test. Such
significance testing considers the question:
E.6.8 For case (b) when a correction of R has to be explicitly included in the calculation of
the corrected result, i.e.
c
c corr = ------ … Eq [1]
R
where c is the raw results with an uncertainty uc, it is obvious that we must include uR in
the uncertainty budget. This led us to a combined uncertainty ucorr on the corrected
result given by:
2 2
ucorr
ccorr
[ ucc ] [ uR
R
] … Eq [2]
But, how are we going to calculate the uncertainty of recovery, uR and uncertainty of
bias, uB?
E.6.9 Example 1:
In a measurement of trace metal analysis in water, the laboratory has shown that a
recovery mean for copper contents in 10 samples of distilled water (with added 3.50
mg/L Copper) is 3.65 mg/L with a standard deviation of 0.18 mg/L. The ‘native’
copper content in the distilled water has been found to be below its detection limit.
µ = x ± t (s/ n ) … Eq [3]
In fact, the uncertainty of recovery, uR as shown by the above equation is the factor,
s/ n .
Hence,
t µ - x / (s/ n )
= 2.64, which is larger than the critical t- value of 2.26 for 9 degree of freedom at
95% confidence.
This indicates that the figure 3.65% found by the experiment is significantly larger
than the 3.50% spiked amount and therefore a bias exists. The experimental value
can be corrected by applying a correction factor which is to be the reverse of the
recovery mean of 104%, i.e. a factor of 0.96.
However, if for some practical reason, the bias cannot be corrected, then the
standard uncertainty due to bias, u(B), is to be calculated by incorporating:
where: R = mean ratio of the recovered data divided by the known value
k = the Student t-value at a given degree of freedom
uR = standard uncertainty of the mean ratio R
R = 1.04
t = 2.64
u R = 0.18 / 10 = 0.057
Hence,
2
u(B ) = {[(1- 1.04 ) / 2.64] (0.057 )2} = 0.059.
E.6.10 Example 2:
An estimate of the method bias can be obtained from QC data by comparing testing
results with the target value. For example, the target value for a Cu check solution is
10.03 ppb. The last 10 days’ testing results are 9.98, 10.33, 10.21, 10.15, 10.23,
10.29, 10.31, 10.27, 10.20, 10.28 ppb. The ratios of laboratory-result over target
value are calculated as 0.9950, 1.0299, 1.0179, 1.0120, 1.0199, 1.0259, 1.0279,
1.0239, 1.0169, and 1.0249. The mean (R) of these ratios is R = 1.01942 with a
standard deviation (SR) of SR = 0.01019. The standard uncertainty is calculated as
the standard deviation of the mean: u(R) = SR/ n = 0.01019/ 10 = 0.00322.
1− R 1 − 1.01942
t= = = 6.03
u ( R) 0.00322
The two-tailed critical t value at 95% for d.f.= n-1 = 9 is 2.26. Therefore, R is
significantly different from 1, i.e. bias exists.
Technical Guide 2, Mar 08 55
It is a general requirement of ISO GUM that corrections should be made for all
significant bias. Thus, a correction factor equal to 1/R shall be applied in the
mathematical model when calculating the testing results for samples. With this factor
applied, u(R) will be included in the calculation of combined standard uncertainty.
If the bias in the above case has not been corrected due to some practical reason,
then, standard uncertainty due to bias u(B) is calculated as u(B)=
{[(1- R)/k] 2 u 2 R } , where R is the mean ratio obtained above, u(r) is the standard
uncertainty of the mean ratio obtained above, and k is the student t-value at a given
degrees of freedom. For the case above, standard uncertainty due to bias without
2
bias correction is u(B) = {[(1- R)/k] u2 R } = {[(1-1.01942)/2.26]2+0.003222} =
0.00918.
Assuming there is no bias detected in the above case and the relative standard
uncertainty of the check solution is u(std) = 0.0022, then u(B) = [u2(std)+u2(R)] =
[0.00222+0.003222] = 0.0039.
Recovery data can also be used to determine bias present in a method. The target
value for recovery is 100%. For example, for measurement of pesticide in butter, a
study has shown a recovery mean for 33 samples is 109% with a standard deviation
of 12%. The standard uncertainty of the mean is u(Rec) = 12%/ 33 = 0.020889.
The t test result is t= 109%-100% /0.020889 = 4.31. This t value is larger than the
critical t value for 32 d.f. at 95% confidence. Therefore, bias exists and has to be
corrected by applying a correction factor which is the reverse of the recovery mean.
WORKED EXAMPLES
F.1.1 METHOD
F.1.1.1 Ascertain the 0.1M sodium hydroxide (NaOH) solution by titrating against 0.1M
potassium hydrogen phthalate (KHP) solution;
F.1.1.2 Then determine the concentration of approximately 0.1M HCl solution by titrating
against the 0.1M NaOH solution.
F.1.2 PROCEDURE
Step 6 : Titrate the NaOH solution against the KHP solution from a 50-ml burette
Step 9 : Titrate the standardized NaOH solution against the HCl solution from a 50-ml
burette
F.1.3 CAUSE AND EFFECT DIAGRAM (for determining the concentration of NaOH
solution)
c(NaOH)
Vol(KHP) MW(KHP)
F.1.4.1.1 Workings:
Mass(KHP)
calibration calibration sensitivity
sensitivity repeatability linearity
Purity (KHP) repeatability
Linearity Mass(gross)
Mass(Tare)
c(NaOH)
Vol(KHP) MW(KHP)
The laboratory QA report shows that a standard deviation of 0.09 mg was found
in this balance used after 7 replicated weighings.
repeatability
calibration bias repeatability
Temperature
End-point
Vol(KHP) MW(KHP)
F.1.4.2.1 Workings:
The manufacturer’s catalog states that the 250-ml flask comes with an
uncertainty of ± 0.15 ml without mentioning the degree of confidence. Hence, a
rectangular distribution of errors is assumed with a factor of 3 . The standard
deviation of the volume is therefore 0.15 / 3 = 0.087 ml
The standard deviation for the variation of the total volume of the volumetric flask
calculated after a series of replicated (10 times) filling and weighing of water to
the mark is found to be 0.014 ml. This value will be used in the later calculation
of the uncertainty in volume measurement.
Take the coefficient of volume expansion for water as 2.1 x 10-4 per °C and the
temperature variation between that of the solution and the calibration temperature as
5 degrees. For the volume of 250 ml used, this will give a 95% confidence interval of:
F.1.4.3.1 Workings:
Where,
Purity of KHP has been provided by the supplier as 99.8% ± 0.2%, meaning P is
0.998 ± 0.002. As there is no confidence level stated for the uncertainty, we
have to take a rectangular distribution of error with a factor of 3 , giving u(P) as:
2 2 2 2
u(MW) = (0.0046 0.0002 0.00068 0.000058 )
= 0.0047
The standard uncertainty in the concentration (Mk) of this KHP standard solution
therefore is expressed as:
u(Mk) 2 2 2
(0.0000205 0.001202 0.00064 2 0.000023 = 0.0014 mol L-1
Mk
Now, the concentration of KHP solution, Mk, is calculated from equation [1] as follows:
Hence, the standard uncertainty u(Mk) in the concentration of KHP solution is:
The concentration of KHP solution is therefore 0.0992 mol L-1 with a standard
uncertainty of 0.00014 mol L-1.
The pipette manufacturer states that the pipette used has an uncertainty of
±0.03ml. By approximating to a rectangular distribution because of unknown
confidence level, the standard deviation of the volume of this pipette to be
measured is 0.03 / 3 or 0.017 ml.
Combining all these 3 sources of uncertainty, we have the uncertainty u(VS) in the
volume transfer of NaOH solution as the square root of the sum of squares of these 3
standard deviations, giving the result of:
2 2 2
u(VS) = (0.017 0.0010 0.013 ) or 0.021
F.1.4.6 Step 6: Titration of the NaOH solution against the standard KHP solution (Va)
The 25 ml NaOH solution is titrated against the standard KHP solution from a 50-ml
burette. Again, we need to consider the sources of uncertainty from the point of view
of the similar 3 factors discussed earlier, via:
The manufacturer states that the burette used has an uncertainty of ± 0.05ml. By
approximating to a rectangular distribution because of unknown confidence level,
the standard deviation of the volume of this pipette to be measured is 0.05 / 3
or 0.029 ml.
In this titration exercise, 25.20 ml of the KHP solution was found to be used in
achieving the end point with the NaOH solution. Hence, using the figures obtained
above, the combined standard uncertainty u(Va) is calculated as :
2 2 2
u(Va) = (0.029 0.012 0.013 ) or 0.034 ml
Where,
Ms = concentration of the NaOH solution
Mk = concentration of the KHP standard solution
Va = Volume of the KHP standard solution used
Vs = Volume of the NaOH solution pipetted for titration
The standard uncertainty in the concentration (Ms) of this NaOH solution therefore is
expressed as :
u(Ms) 2 2 2
(0.00141 0.00135 0.00084 ) = 0.0021 mol L-1
Ms
Now, the concentration of NaOH solution, Ms, is calculated from equation [2] as
follows:
Hence, the standard uncertainty u(Ms) in the concentration of NaOH solution is:
As 25-ml volume of the standardized NaOH solution is used for titration, similar
considerations can be applied as in Step 5, giving 25 ml volume with a standard
uncertainty u(Vb) of ± 0.021 ml.
As in step 6, the HCl solution is titrating the 25-ml NaOH solution from a 50-ml
burette. The combined standard uncertainty, u(Vc) is therefore the same as 0.034
ml under similar assumptions.
where,
Mh = concentration of the HCl solution
Ms = concentration of the NaOH standard solution
Vb = Volume of the NaOH standard solution used
Vc = Volume of the HCl solution from burette for titration
The relative standard uncertainty in the concentration (Mh) of this HCl solution
therefore is expressed as:
u(Mh) 2 2 2
(0.00213 0.00135 0.00084 = 0.0027 mol L-1
Mh
Now, the concentration of HCl solution, Mh, is calculated from equation [3] as follows:
Hence, the standard combined uncertainty u(Mh) in the concentration of HCl solution
is:
F.1.5 Remarks:
In this acid/base titrimetry, the followings are its possible sources of error and some
of them have been taken into account in this example. If additional sources of error
were to be considered significant, they would be considered too:
F.2.1 Specification
b) The sample is extracted in the presence of NH4OH with alcohol, ethyl ether, and
petroleum ether.
c) Fatty acid methyl esters of extracted fat are prepared by using NaOCH3 and
BF3 CH3OH esterification.
e) The residue is then dissolved in heptane and the solution obtained is transferred
to 25 mL volumetric flask and topped up with heptane to the mark.
With recovery rate R available, concentration of linoleic acid in the milk powder
sample (Cspl) can be calculated by:
whereby,
Cspl: Concentration of linoleic acid in the milk powder sample (in mg/g).
Cml: Concentration obtained from calibration curve for methyl linoleate (in mg/mL).
V: Final volume of the solution before injection (25 mL).
Fla: Formula weight of linoleic acid (in g/mol).
Fml: Formula weight of methyl linoleate (in g/mol).
Wspl: Milk Powder sample weight (in g).
Cspl
Wspl
Recovery (R)
Cml: The Cml result obtained from calibration curve for the sample was 7.15 mg/mL.
The relationship for forced zero least square fitting is y = bx with b being the slope of
the calibration curve:
b=
∑x yi i
∑x 2
The concentration xobs of the analyte from a sample which produces an observed
response yobs is then given by xobs=yobs/b. The uncertainty u(xobs,y) in a predicted
value xobs due to variability in y can be estimated from the variance of residuals S as
for Generic Example E.4 above:
x y xy x2
0 0 0 0
1 133 133 1
2 278 556 4
4 558 2232 16
10 1192 11920 100
Sum 17 2161 14841 121
Thus, b =
∑x y i i
= 14841/121 = 122.6528926 Î y = 122.6528926x. Therefore:
∑x 2
x y Calculated yc (y-yc)2
0 0 0 0
1 133 122.652893 107.06263
2 278 245.305785 1068.9117
4 558 490.61157 4541.2005
10 1192 1226.52893 1192.2467
Thus,
S2 = (yi-yc)2/(n-2) = (107.06263+1068.9117+4541.2005+1192.2467)/(5-2)
= 2303.14.
V: The volumetric glassware used for topping the final solution to 25 mL has a
certified value of 25.040 ± 0.015 mL at 20 C, obtained from supplier’s specification.
For this study, a rectangular distribution has been chosen. Therefore the uncertainty
due to calibration is 0.015/ 3 = 0.00866 mL.
2 2
Therefore, u(V) = (0.00866 0.0210 ) = 0.0227.
Fla: As for Working Example F1, molecular weight (MW) of linoleic acid (C18H32O2)
and its uncertainty are calculated as:
C H O
Value 12.0107 1.00794 15.9994
Uncertainty 0.00046 0.00004 0.00017
C H O
Value 12.0107 1.00794 15.9994
Uncertainty 0.00046 0.00004 0.00017
Recovery (R): During another previous study on a similar sample, repeated recovery
tests were done on a single sample and found to have an average of 91.3% recovery
with a standard deviation of 5.4%. Thus, the relative standard deviation is
5.4%/91.3%=0.0591. For current sample, recovery was 0.950 (i.e. 95.0%). Thus, the
standard uncertainty due to recovery is 95%x0.0591=5.61%.
Standard uncertainties due to Cml, V, Fla, R, Fml and W spl are combined first by the
spreadsheet method:
The above standard uncertainty obtained is then combined with that due to precision
to give total combined standard uncertainty. As precision obtained above is relative
standard deviation, thus,
The concentration of linoleic acid in the tested milk powder has been found to be:
F.2.5 Comments:
According to the law of propagation of errors, for addition and subtraction relationship,
e.g. x = x1+x2, the combined standard uncertainty is the square root of the sum of
square of uncertainty of individual components, i.e. u(x) = [u2(x1)+u2(x2)]. Thus,
contribution of each component to the combined standard uncertainty can be
compared directly between u(x1) and u(x2).
2 2
u(x)/x= {[u(x1)/x1] [u(x2)/x2] }
For the working example discussed above, the relationship is of multiplication and of
division. Thus, to compare each component’s contribution, each component
uncertainty shall be converted to relative standard uncertainty.
From the table above, it is obvious that the main contributions are from Cml, R and r.
In fact, one can work out the combined standard uncertainty from these three
sources only by ignoring other minor sources. In this case, u(Cml)/Cml= (0.05472 +
0.05912 + 0.1702) = 0.188. Thus, u(Cml)=0.188 x Cml = 0.188 x 17.90667 = 3.37 mg/g.
Thus, it is obvious that the uncertainty calculated from only major sources is not
significantly different from that obtained after considering all sources.
F.3.1 Method
CKOH W(oil)
MW(KOH)
Acid Value
Titration (V)
The volumetric flask used for the solution preparation has the volume 1000 mL ±
0.4 mL at 20oC. The appropriate standard deviation of the calibrated volume
using a rectangular distribution is 0.4/ 3 or 0.23 mL.
Since the actual temperature and the flask calibration temperature is -3 oC with
95% confidence, at volume coefficient of water expansion 2.1 x 10-4 per oC per
mL, the possible volume variation is 1000 x 3 x 2.1 x 10-4 or 0.63 mL. The
corresponding standard deviation is 0.63 /1.96 or 0.32 mL.
The standard deviation of the flask filling is less than 1/3 of the standard deviations
for calibration and temperature variation, and is thus neglected.
2 2
u(V) / V = (0.23 0.32 ) / 1000
= 0.00039.
The concentration of HCl is mHCl / MHCl.V where MHCl is the molecular weight of HCl.
The manufacturer of the HCl solution indicates a possible deviation of its titer of
0.02% per oC. Taking a possible temperature difference in the manufacturer' s
laboratory of -2oC (with 95% confidence), the standard uncertainty of mHCl is:
2 2
u(Cst)/Cst = (0.00039 0.00022 = 0.00045.
The exact concentration of the KOH solution is established before its use by titration
against the standardized HCl solution.
Where,
Vst is the volume (mL) of the standard HCl solution used for titration of the volume
VKOH (mL) of the KOH solution.
For transfer of an aliquot of the KOH solution to the conical flask, a glass pipette
of volume 5 ± 0.01 mL is used. Taking a possible temperature variation of ± 3oC
with 95% confidence, and repeatability of filling the pipette (standard deviation)
0.0033 mL, one can calculate u(VKOH)/VKOH = 0.0015
The possible temperature variation is the same as that mentioned above, the
standard deviation of filling is 0.0033 mL, and the standard deviation of end point
detection arising due to the drop size of the burette (0.017 mL) is 0.0098 mL.
Thus, the maximum value of u(Vst)/Vst = 0.013 if CKOH = 0.1 mol/L , and the
corresponding Vst = 1mL.
The test method recommends the use of KOH molecular weight of 56.1 Instead
of the complete value MKOH = 56.10564; hence, in this case,
It is clear that the uncertainties of the molecular weight of KOH and weighing of
oil sample are negligible. Hence,
2 2
u(AV)/AV = [u(VKOH)/VKOH] [u(CKOH)/CKOH] = 0.018
NOTE: The detection of the end point of the titration is a dominant source of
uncertainty. If a commercial burette, for example, has a drop size of 0.043
mL, the expanded uncertainty will increase to 0.07.
Moreover, the colour of the oils and the possible change in the indicator
behaviour near the end point in the oil-solvent mixture are not taken into
consideration. The same relates also to the influence of atmospheric CO2 on
CKOH.
The time is measured for a fixed volume of liquid to flow under gravity through the
capillary of calibrated viscometer under a reproducible driving head and at a closely
controlled and known temperature. The kinematic viscosity is calculated as the
product of the measured flow time and the calibration constant of the viscometer.
F.4.4 Example
Upon analysis of a fuel oil sample, the following results were obtained:
Therefore,
Kinematic viscosity of sample found = 177.8 mm2/sec
Combined standard uncertainty = 0.301 mm2/sec
Expanded uncertainty with a coverage factor of 2 = 0.60 mm2/sec
Crude fibre is defined in the method scope as the amount of fat-free organic
substances which are insoluble in acid and alkaline media. There is no suitable
reference material available for this method. However, both collaborative inter-
laboratory/proficiency studies (repeatability and reproducibility) and in-house
repeatability studies have been carried out to evaluate method performance.
During the analysis, the sample is treated to digest most components, leaving behind
all the undigested material. The test method requires blank correction to be done.
The percentage of weight loss after blank correction is defined as the "fibre content"
by the method. Thus, the fiber content as a percentage of the sample by weight, Cfiber,
is given by:
whereby,
a: original sample weight
b: weight loss for the sample
c: weight loss for the blank (crucible)
F.5.3.1 Data from collaborative studies and from in-house repeatability studies
Five different feeding stuffs representing typical fibre and fat concentrations were
analysed during the studies. Participants in the studies carried out all stages of the
method, including grinding of the samples. The repeatability (sr) and reproducibility
(sR) estimates obtained from the studies are presented in the table below.
During the in-house repeatability studies, experiments were also done to evaluate the
repeatability (within batch precision) for the feeding stuffs at the similar concentration
as those for collaborative studies. The results are presented in the last of the column
of the table below.
Fiber Content (%, w/w)
Sample Collaborative Interlaboratory/Proficiency Studies In-House
Repeatability
Mean Reproducibility Repeatability Standard
Standard Deviation
Standard Deviation (sR) Deviation (sr)
A 2.3 0.293 0.198 0.193
B 12.1 0.563 0.358 0.312
C 5.4 0.390 0.264 0.259
D 3.4 0.347 0.232 0.213
E 10.1 0.575 0.391 0.327
To complete the uncertainty calculation we need to consider whether there are any
other effects not covered by the collaborative studies which need to be addressed.
The collaborative studies covered different sample matrices and the pre-treatment of
samples, as the participants were supplied with samples which required grinding
prior to analysis. The uncertainties associated with matrix effects and sample pre-
treatment do not therefore require any additional consideration.
However, repeatability data from collaborative studies and from in-house studies
does not reveal individual participating laboratory’s bias. This bias should be
evaluated separately to determine if it is significant compared to the reproducibility
standard deviation.
For this particular laboratory, the “constant weight” was achieved within 2mg only.
The uncertainty from this bias is thus 0.002/ 3 = 0.00115g. As the method specified
a 1g sample to be used, the standard uncertainty due to weighing bias is thus
0.115%. From the table above, it is obvious that for all fibre concentrations, this
uncertainty is smaller than the reproducibility standard deviation, and for all but the
lowest fibre concentrations is less than 1/3 of the sR value. Again, this source of
uncertainty can usually be neglected. However for low fibre concentrations (e.g. 2.3%
w/w in the table above), this uncertainty is more than 1/3 of the sR value so an
additional term should be included in the uncertainty calculation for such a low level
sample.
For 3% w/w or above level, sR can be used as the standard uncertainty. E.g, for 3.4%
w/w level, the standard uncertainty is 0.347% and the expanded uncertainty with k=2
(95% confidence) is 0.69%. For those below 3% w/w level, e.g. 2.3%, the combined
standard uncertainty is (0.115%2+0.293%2) = 0.31%. The expanded uncertainty is
thus 0.62% (k=2 at 95% confidence level).
Sensitivity Sensitivity
Linearity Linearity
Weight (gross)
Environment
Environment
Instrument effect
Instrument effect
Precision
Precision MOISTURE
Calibration Calibration
Sensitivity Sensitivity
Linearity Linearity
Environment (
Environment
Instrument effect
Instrument effect
Precision
Precision
Calibration
F.6.3 Purpose
b. Weighing records:
Before drying:
The external calibration of the balance used states that the difference from the
actual weight on the scale pan and the reading on the scale is within ± 0.5 mg with a
95% confidence.
0.5 or 0.255 mg
1.96
NOTE: This component uncertainty has to be taken into account twice because of
two weighings involved each time. Two sets of weighings were made, one
of sample before drying and one of sample after drying.
F.6.4.3. Sensitivity
The expanded uncertainty using a coverage factor of 2 (to get 95% confidence limit) is:
U(moisture) = 0.0041 % x 2
= 0.0082 %
Therefore, the result is 29.52 % ± 0.0082 % with approximately 95% confidence level.
CxDxV100
Benzoic Acid (ppm) (w/w):
S
C V100 S D
C
For this analysis, 4 levels of calibration standards were prepared:
0, 10, 20, 40 and 80 ml/g
The 4 calibration standards and their blank were measured and the following results
were obtained:
X Conc’n y Area
xy x2 Calculated yc (y-yc)2
(mg/l) (mAU’s)
0 0 0 0 0 0.000
10 231.0 2310 100 235.054 16.436
20 462.5 9250 400 470.108 57.885
40 945.7 37828 1600 940.216 30.069
80 1880.1 150408 6400 1880.433 0.111
Sum 3519.3 199796 8500 104.501
Slope (b) = Sum xy / Sum X2 23.505
S2 = sum (y-yc)2 / (n-2) 34.834
Var (x) = S2 / b2 0.063047
u(C) = sqrt(S2 / b2) 0.251091
u(C) = 0.251091 mg/l
S
Use 2-decimal place balance for weighing of sample
Test 1 Test 2
Glassware ID: 159 158
u(100), µl = 80.11 96.89
u(100), ml = 0.08011 0.09689
D
If no dilution is required:
Pipette and volumetric volume are set as 1, standard uncertainty are set as zero.
If dilution is required:
After extraction, pipette a suitable amount of sample extract into a volumetric flask
and make up volume with 70% ethanol. Obtain combined standard uncertainty from
calibration record for glassware.**
Test 1 Test 2
Pipette Vol, Vol Flask, Pipette Vol,
Vol Flask, ml
ml ml ml
Value 10 50 10 50
Std Uncertainty 0.01677 0.02140 0.00779 0.03479
C.V. 0.001677 0.000428 0.000779 0.000696
C.V. sq 0.000003 0.000000 0.000001 0.000000
Sum of C.V. sq 0.000003 0.000001
Sqrt (Sum of C.V. sq) 0.001732 =u(D)/(D) =A 0.001000 =u(D)/(D) =A
Dilution factor (D) 5 =B 5 =B
u(D) 0.008660 = A*B 0.005000 = A*B
C D V100 S
Test 1 Value 21.010 5 100 10.00
Std Uncertainty 0.251091 0.008660 0.080110 0.013900
C.V. 0.011951 0.001732 0.000801 0.001390
C.V. sq 0.000143 0.000003 0.000001 0.000002
Sum of C.V. sq 0.000149
Sqrt (Sum of C.V. sq) 0.012207 = u(Benzoic Acid)/(Benzoic Acid) = A
Benzoic Acid, mg/kg 1050.50 = (C*D*V100) / S = B
u(Benzoic Acid), mg/kg 12.822987 = A*B
Purity of Standard
Lot: 14141
Purity: 100 ± 0.5%
As there is no confidence limit of the purity, we take the quoted uncertainty as the
rectangular distribution.
The test result takes the form of: 1051.63 ± 26 ppm (w/w)
** calibration records not shown in details herein. However, working examples of such
calibration have been shown in this Guide elsewhere.
F.8.1. Procedure
iii Obtain the absorbance readings of standards at 570 nm and plot the
calibration curve.
b. Sample preparation
i If the sample contains residual chloride, add 1 drop of NaAsO2 solution per
0.1 mg residual chlorine.
iii Adjust temperature to be similar to those used for obtaining the standard
curve.
F.8.2. Calculation
A
mg F- /L = x 50 x dilution factor of sample, if any
ml sample
where
a. Stock fluoride solution: 221 mg (±0.40 mg) NaF (equivalent to 100 mg F) in 100
ml (±0.23 ml) distilled water (i.e. 1000 mg F-/L)
Std uncertainty u(w) = 0.40/ 3 = 0.2309
Std uncertainty u(v) = 0.23/2 = 0.115 ml
b.1 Intermediate fluoride standard solution: Dilute 1.00 ml (±0.03 mL) stock standard
to 100.0 ml (±0.23 mL) solution to give 10 mg F-/L
u ( Dilution factor )
= (0.015 / 1.00) 2 + (0.115 / 100) 2
100
b.2 Working fluoride standard solution: Dilute 10.0 ml (±0.08 mL) intermediate F
standard to 100.0 ml (±0.23 mL) solution to give 1.0 mg F-/L
u( Dilution factor )
= (0.04 / 10.0) 2 + (0.115 / 100) 2
10
Upon analysis, the sample absorbance = 0.2531, which gave 0.424 mg/L with
standard uncertainty of 0.0405mg/L as obtained from the calibration curve.
0.424 x 50
The calculated Fluoride F- content in sample = ---------------- = 0.424mg/L
50
= 0.0967
F.8.5. Results
The fluoride content of the sample = 0.42 mg/l ± 0.08 mg/l with a coverage factor of 2
(95% confidence)
Ts
Vs = Kp * Cp * ( Dp )
(Ps * Ms)
Kp Cp Dp
Constant
Stack
Velocity
Ts Ps Ms
a. Dp – Differential pressure
b. Ts – Temperature of stack
Standard uncertainty of temperature probe of the instrument at 39.9oC = 0.37oC,
based on the calibration of the probe against the working thermocouple.
Given data:
Kp = 34.97
Cp = 0.84
Ts = (273 + 39.9) K = 312.9 K
Ps = 759.4 mm Hg
Ms = 29.9 g/g-mole
Dp = 26.0 mm Water
F.9.6. Reporting
Velocity of Stack Gas = 17.52 ± 0.99 with a coverage factor of 2 (95% confidence)
F.10.1 Purpose
To evaluate the measurement uncertainty of 60 ppm of V
F.10.2 Procedure
1.1 20-50 gm (W) of sample was pretreated in a clean platinum dish and the residue was
made up to volume with water in a 100-mL volumetric flask.
1.2 The stock solution was prepared from a 1000 mg/L commercial standard. 5, 10, 25
and 50 mg/L V standards were prepared from the 250 mg/L stock solution
1.3 The concentration of V in the sample (Cv, mg/L) was obtained from the calibration
curve.
1.4 The concentration (C, mg/L) in the original sample was calculated with the following
formula.
V x CV
C = ------------ … [1]
W
Mass (sample)
Sensitivity
5 mg/L Std
Repeatability
10 mg/L Std
50 mg/L Std
V content
The weight of the sample was obtained by subtracting the weight of empty Pt dish from
weight of Pt dish and sample.
The calibration certificate states that at 95% confidence level, a weight obtained
by difference within the same range was within ± 0.2 g. This uncertainty
component can be expressed as a standard deviation by dividing 0.2 by 1.96,
giving 0.102 g.
Sources of uncertainties:
The manufacturer states that for the 100 ml volumetric flasks, the error was ± 0.1 mL
at 20oC without stating the confidence level.
10 fill and weigh exercises on the 100-mL volumetric flask gave the standard
deviation as 0.03332 mL.
c. Temperature effect
Using a rectangular distribution, the standard deviation for temperature variation was
0.105 / 3 or ± 0.0606 mL.
The manufacturer states that for the 25 mL pipette, the error was ± 0.03 mL at a
temperature of 20oC. No confidence level was stated.
10 fill and weigh exercises on the 25 mL volumetric flask gave the standard deviation
as 0.01253 mL.
c. Temperature effect
Using a rectangular distribution, the standard deviation for temperature variation was
0.026 / 3 or ± 0.01516 mL.
Hence, using the figures obtained above, the combined standard uncertainty u(V1) was:
Concentration of stock solution (1000 mg/L) has been provided by the supplier as 1012 mg/L.
As there is no confidence level stated for the uncertainty, we ignore its uncertainty.
Standard uncertainty in the concentration (C1) of this standard solution (25 mg/L) was,
u(C1)
----------- = ! (0.000913 2 + 0.00080 2) = 0.0012
C1
The concentration of this solution, C1, was 253 mg/L. Hence, the standard uncertainty u(C1)
was u(C1) = 0.0012 x 253 = 0.310 (mg/L)
V5 x C1
C5 (mg/L) = --------------
V
The manufacturer states that for the 2 mL pipette the error was # 0.006 mL at a
temperature of 20oC without stating the confidence level.
8 fill and weigh exercises on the 2-mL pipette gave the standard deviation as
0.00055 ml.
c. Temperature effect
Taking the temperature variation of 5 oC and the coefficient of volume expansion of
glass as 2.1 x 10-4 per oC, the 95% confidence level of volume measurement was,
Using a rectangular distribution, the standard deviation for temperature variation was
0.0021 / 3 or ± 0.00121 mL.
Hence, using the figures obtained above, the combined standard uncertainty u(V5) was,
u(C5)
----------- = ! (0.0014 2 + 0.00080 2 + 0.0012 2) = 0.0016
C5
The concentration of this solution, C5, was 5.06 mg/L. Hence, the standard uncertainty u(C5 )
was u(C5 ) = 0.0016 x 5.06 = 0.008 mg/L.
Step 6 Prepare V standard solution C10 (10 mg/L), by diluting 4 mL of C1(250 mg/L)
to 100 mL with water
The procedure was the same as Step 5 and the uncertainties were summarized below:
u(C10)
----------- = ! (0.00552 + 0.000802 + 0.00122) = 0.0056
C10
The concentration of this solution, C10 was 10.12 mg/L. Hence, the standard uncertainty
u(C10 ) was u(C10 ) = 0.0056 x 10.12 = 0.057 mg/L.
Step 7: Prepare V standard solution C25 (25 mg/l), by pipetting 10 mL of C1(250 mg/L)
to 100 mL with water
The procedure was the same as Step 5 and the uncertainties were summarized below:
u(C25 )
----------- = ! (0.0016 2 + 0.00080 2 + 0.00122) = 0.0018
C25
The concentration of this solution, C25, was 25 mg/L. Hence, the standard uncertainty u(C25 )
was u(C25 ) = 0.0018 x 25.3 = 0.046 mg/L.
The procedure was the same as Step 5 and the uncertainties were summarized below:
u(C50 )
----------- = ! (0.00096 2 + 0.00080 2 + 0.00122) = 0.0013
C50
Now, the concentration of this solution, C50 , is 50 mg/L. Hence, the standard uncertainty
u(C50 ) is u(C50 ) = 0.0013 x 50.6 = 0.063 mg/L.
The least squares method was used to obtain the relationship between calibration data pairs
(xi, yi).
There were four main sources of uncertainty to consider when estimating uncertainty of CAl.
Of these four sources, the most significant is A. Method for estimating A introduced below
was through variance of residuals, S.
S2 = $ (yi- yc)2/(n-2)
where
yi is reading of ith calibration point and yc is the calculated reading from the relation y = a + bx
while n is the number of calibration points. In IP501, n = 5.
Concentration, xi Response, yi
0 0.4088
5.06 36.58
10.12 71.96
25.30 180.7
50.60 354.7
− −
b=
∑x y − n x y
i i
− −
a = y− b x
−2
∑x − n x
2
i
In this analysis,
x Y xy x2
0 0.4088 0.0 0.0
5.06 36.58 185.1 25.6
10.12 71.96 728.2 102.4
25.30 180.7 4571.7 640.1
50.60 354.7 17947.8 2560.4
Sum 91.08 644.349 23432.9 3328
Therefore,
− −
b=
∑ xi yi − n x y =
4686.6 − 5 ×18.216×128.87
= 7.006
−2 3228− 5 ×18.2162
∑x 2
i −nx
− −
a = y− b x = 128.87-7.006 x 18.216 = 1.250
X Y Calculated yc (y-yc)2
0 0.4088 1.24952 0.7068126
Sum 6.701
Now, V concentration of this solution, C, was 60.3 ppm. Hence, the standard uncertainty u(C)
is u(C) = 60.3 x 0.0088 = 0.53 (ppm)
Hence, V concentration in the sample analysed was found to be 60.3 # 1.1 (ppm)
1. APHA standard method for total coliform count of reservoir water samples by membrane
filtration was used. The coliform group is defined as those facultative anaerobic, gram-
negative, non-spore-forming, rod-shaped bacteria that develop red colonies with a
metallic sheen within 24 hours at 35 0C on an Endo-type medium containing lactose. The
standard method specifies incubation time of 22 to 24 hours. As coliform counts are
time-sensitive, an incubation time of 22 hours, and M-Endo Medium were used in this
MU determination.
Receipt of sample
Sub-sampling
(mixing, and measuring volume of sample using pipettor)
Preparation of dilutions
Filtration
5. 21 samples were set up and analysed on different days in duplicate pairs, analysts on
different days, using different equipment (incubators, pipettors) on different days and
using different batches of reagents and media on different test days.
n
RSDR = [ [(log ai-log bi)/xi]2] / 2n
i=1
where
(log ai-log bi)/xi = the relative difference between the duplicate logarithmic results
i =1,2,… .n
Diff /
TC / Filter TC / Filter Mean, D (Diff /
Log ai Log bi 2
Duplicate 1 Duplicate 2 Mean Difference (log ai -log Mean)
2
Sample No. (ai) (bi) xi log ai - log bi bi)/xi D
1 10 13 1.0000 1.1139 1.0570 -0.1139 -0.1078 0.0116
2 22 23 1.3424 1.3617 1.3521 -0.0193 -0.0143 0.0002
3 28 25 1.4472 1.3979 1.4225 0.0492 0.0346 0.0012
4 41 55 1.6128 1.7404 1.6766 -0.1276 -0.0761 0.0058
5 52 49 1.7160 1.6902 1.7031 0.0258 0.0152 0.0002
6 46 35 1.6628 1.5441 1.6034 0.1187 0.0740 0.0055
7 44 31 1.6435 1.4914 1.5674 0.1521 0.0970 0.0094
8 29 41 1.4624 1.6128 1.5376 -0.1504 -0.0978 0.0096
9 64 50 1.8062 1.6990 1.7526 0.1072 0.0612 0.0037
10 23 21 1.3617 1.3222 1.3420 0.0395 0.0294 0.0009
11 22 19 1.3424 1.2788 1.3106 0.0637 0.0486 0.0024
12 30 42 1.4771 1.6232 1.5502 -0.1461 -0.0943 0.0089
13 37 33 1.5682 1.5185 1.5434 0.0497 0.0322 0.0010
14 18 22 1.2553 1.3424 1.2988 -0.0872 -0.0671 0.0045
15 23 18 1.3617 1.2553 1.3085 0.1065 0.0814 0.0066
16 45 31 1.6532 1.4914 1.5723 0.1619 0.1029 0.0106
17 25 36 1.3979 1.5563 1.4771 -0.1584 -0.1072 0.0115
18 13 11 1.1139 1.0414 1.0777 0.0726 0.0673 0.0045
19 15 19 1.1761 1.2788 1.2274 -0.1027 -0.0836 0.0070
20 66 69 1.8195 1.8388 1.8292 -0.0193 -0.0106 0.0001
21 7 30 0.8451 1.4771 1.1611 -0.6320 -0.5443 0.2963
Step 1. Transform the raw data by taking the log10 of the data (Column 4, 5).
Step 3. Calculate the difference between the transformed replicates (Column 7).
Step 4. Divide the difference between the transformed replicates by the Mean (Column 8).
Step 6. Add the differences together (Column 9) and divide by 2n, where n = the total
number of pairs of duplicates (for this example n = 21) to get 0.009560
Step 7. Take the square root of the results in Step 6; this equals the Relative Standard
Deviation of Reproducibility, which is 0.0978.
7. On examination of the data set, the relative error, difference / mean of sample No. 21
was 0.5443 which is much larger than the values obtained from other duplicate pairs.
The result of this sample appeared to be outlier. The standard Grubbs test for
identification of outliers of duplicate pairs was used,
Where:
RD = Relative Difference for each duplicate pair, as a decimal fraction [RD = (ai-bi) /
X mean]
The calculated T value, 3.94, of the result of sample no. 21 exceeded the critical value of
2.58 for 21 data points of the Grubbs T value table at 5% risk of false rejection. Hence,
the result of this sample was found to be an outlier. On investigation, the pipettors used
for this test were defective. The result was excluded from the calculation of the
measurement uncertainty of the determinations. The measurement uncertainty was re-
calculated based on 20 sample determinations as follows:
Table 2 : Total Coliform (TC) Tests (APHA Method) on a Series of Different Water
Samples from Reservoir
TC / Filter TC / Filter (Diff /
Log ai Log bi
Sample Duplicate 1 Duplicate 2 Difference Diff / Mean, D Mean)2
No. (ai) (bi) Mean xi log ai - log bi (log ai -log bi)/xi D2 Technician
1 10 13 1.0000 1.1139 1.0570 -0.1139 -0.1078 0.0116 A
2 22 23 1.3424 1.3617 1.3521 -0.0193 -0.0143 0.0002 B
3 28 25 1.4472 1.3979 1.4225 0.0492 0.0346 0.0012 A
4 41 55 1.6128 1.7404 1.6766 -0.1276 -0.0761 0.0058 B
5 52 49 1.7160 1.6902 1.7031 0.0258 0.0152 0.0002 A
6 46 35 1.6628 1.5441 1.6034 0.1187 0.0740 0.0055 B
7 44 31 1.6435 1.4914 1.5674 0.1521 0.0970 0.0094 A
8 29 41 1.4624 1.6128 1.5376 -0.1504 -0.0978 0.0096 B
9 64 50 1.8062 1.6990 1.7526 0.1072 0.0612 0.0037 A
10 23 21 1.3617 1.3222 1.3420 0.0395 0.0294 0.0009 B
11 22 19 1.3424 1.2788 1.3106 0.0637 0.0486 0.0024 A
12 30 42 1.4771 1.6232 1.5502 -0.1461 -0.0943 0.0089 B
13 37 33 1.5682 1.5185 1.5434 0.0497 0.0322 0.0010 A
14 18 22 1.2553 1.3424 1.2988 -0.0872 -0.0671 0.0045 B
15 23 18 1.3617 1.2553 1.3085 0.1065 0.0814 0.0066 A
16 45 31 1.6532 1.4914 1.5723 0.1619 0.1029 0.0106 B
17 25 36 1.3979 1.5563 1.4771 -0.1584 -0.1072 0.0115 A
18 13 11 1.1139 1.0414 1.0777 0.0726 0.0673 0.0045 B
19 15 19 1.1761 1.2788 1.2274 -0.1027 -0.0836 0.0070 A
20 66 69 1.8195 1.8388 1.8292 -0.0193 -0.0106 0.0001 B
Summation 0.1052
Number of duplicate analysis, n 20
2 x number of duplicate (2 x n) = 40
Sum of (Diff / Mean)2/ 2n = 0.0026311
Relative Standard Deviation, RSD = 0.0513
Coefficient of Variation, CV% = 100 x RSD 5.1295
A sample from the same matrix/source was found to have a total coliform count of 60 cfu
/ filter. The cfu was converted to a log value of 1.7782. Using the RSDR value obtained
from the results listed in the above Table 2, the following expanded uncertainty was
obtained using the following equation:
In this example, 10 ml of sample was filtered. Therefore, the total coliform count per 100
ml sample is 60 x 10 cfu/ 100 ml = 600 cfu/ 100 ml. The measurement uncertainty of the
total coliform count of the sample is reported as:
Total coliform count, cfu/ 100 ml: 6.0 x 102 with confidence interval of 3.9 x 102 to 9.1 x
102.
8. The same approach can be used for the estimation of personal repeatability RSD in the
laboratory. Using the above example, the personal repeatability RSDr of Technician A
can be calculated as follows:
i=n
[(log ai-log bi)/xi]2
Technician A Repeatibility RSDr = i=1
2n
SD
SQD 0.000679 0.001355
Sp
SQD = (4 x 0.000679) + (4 x 0.001355) / (5 + 5 - 2) equals 0.001017
0.0318919
Sp 7
Sp x SQRT (1/5 +
1/5) 0.020170
0.3873719
t= 0.007813 / 0.020170 equals 2
The tabulated 95% critical value is 2.306 for 8 degrees of freedom (i.e. 5 + 5 – 2).
This critical value exceeds the calculated value of 0.3874; therefore H0 is not rejected and
there is no difference between the means of results obtained. There are no significant
differences between the growth promotion performances of the two batches of medium.
The general equation used for the above Student – t statistic for the comparison of two sets
of data is as follow:
t= ( X1 – X2 )
Sp (1/n1 + 1/n2)
Sample storage
before testing Nature of Sample Dilutions
Homogeneity
pH Toxicity
Matrix Diluents
Time
Temperature
Microflora
Volume of dilutions
Pipettor
Calibration
Homogeneity
CFU/ 100ml
Repeatability
Autoclave Time
Competency
Humidity
Recovery
Properties Training
Temperature Air quality
pH Weighing
Calibration
Temperature
Media
Incubation Counting Operator Environment
effects
10. ISO/TS 21748, Guidance for the Use of repeatability, reproducibility and trueness
estimates in measurement uncertainty estimation.
11. ISO/TC 69, Application of Statistical Methods, SC 6, Measurement Methods and Results
12. ISO/TS 19036, Microbiology of Food and Animal feeding stuffs – Guidelines for the
Estimation of Measurement Uncertainty for Quantitative Determinations
13. CCIL Protocol for Estimating Measurement Uncertainty Using QC Data (Type A )
5. Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results
B N Taylor and C E Kuyatt, NIST Technical Note 1297.1003
Technical Guide 2, Mar 08 104